Google rolls out modest improvements to Gemini 2.5 Pro

Google has rolled out another update to its flagship AI model, Gemini 2.5 Pro. The latest version brings modest improvements across a range of benchmarks and maintains top positions on tests like LMArena and WebDevArena. According to Google, the new release also handles challenging programming, knowledge, and logic tasks better than before. The model also "can be more creative with better-formatted responses."

Deepseek's R-1 still offers the best price-performance ratio, but Gemini 2.5 Pro clearly leads among proprietary Western AI models, staying ahead of OpenAI's offerings. | Image: Google

Developers can access the updated Gemini 2.5 Pro in preview through Google AI Studio and Vertex AI. Google plans to promote this preview to a stable, generally available version in the coming weeks. The update is also rolling out to the Gemini app starting today.

Join our community

Join the DECODER community on Discord, Reddit or Twitter - we can't wait to meet you.

Google rolls out modest improvements to Gemini 2.5 Pro

Google releases open-source LMEval to benchmark language and multimodal models

Google expands access to Veo 3, its viral new video model, through the Gemini app

Google shows AI filmmaking tool, XR glasses and launches $250 Gemini subscription

AI agents outperform human teams in hacking competitions

Wait a minute! Researchers say AI's "chains of thought" are not signs of human-like reasoning

Anthropic releases Claude 4 with new safety measures targeting CBRN misuse

Google rolls out modest improvements to Gemini 2.5 Pro

Google releases open-source LMEval to benchmark language and multimodal models

Google expands access to Veo 3, its viral new video model, through the Gemini app

Google shows AI filmmaking tool, XR glasses and launches $250 Gemini subscription